It’s almost as if Google has something to prove after taking a break from launching new cutting-edge AI models earlier this year. Just two weeks after the release of the Gemini 4 Argon, Google announced the Gemini 4 Carbon. This isn’t the equivalent of the Flash model we were expecting, but rather an improvement on Argon, and like that model, Google is working on a phased release that could take a while.
Stop me if you’ve heard this before: Google says its latest AI model reaches a new threshold in coding, agency tasks, and scientific analysis. In fact, it’s so capable that Google can’t let you use it yet.
Gemini 4 Carbon reportedly makes great strides in cybersecurity applications. Google says the model shows a “significant jump” in vulnerability detection, validation and remediation. Carbon is also apparently twice as likely to generate vulnerability-free code compared to Argon. Meanwhile, the Chrome team claims that Carbon is 10 times better at generating “high-quality” patches compared to non-Google models. Google also notes that Carbon has increased its score by 89 points in WebDev Arena, now ranking number four in the rankings.
Google has some numbers to back up these vague assessments (below), which show Carbon at the top of the charts. Google positions Carbon as an entirely new model, but for all we know, it could be a post-training update to Argon. The two models have a lot in common, but the Carbon outperforms the Argon by slim margins in every test Google has run. Their new model also outperforms the GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1 in almost all tests. Benchmarks don’t tell the whole story, of course, and it’s notable that each frontier model scores in the same neighborhood in these tests. The difference between Carbon and the competition is only a few points.
